[SVE] Lower scalable vector mul operations.

This allows us to remove extra patterns from AArch64SVEInstrInfo.td
because we can reuse those required for fixed length vectors.

Differential Revision: https://reviews.llvm.org/D85328